MicroBT Whatsminer M21

The MicroBT Whatsminer M21 is a 2019-generation SHA-256 ASIC rated at 31 TH/s for 1,860 W, or roughly 60 J/TH. It is a legacy machine: too power-hungry to mine Bitcoin profitably at most grid rates, but still genuinely useful as a controllable space heater, a low-stakes learning rig, or a cheap repair-and-run project where electricity is near-free.

MicroBT built the M-series reputation on rugged, serviceable hardware, and the M21 is an honest example of where the company stood at the tail end of the 7nm era. Below we go deeper than the original spec sheet: the silicon and control board inside it, what it actually pulls at the wall, the firmware reality in 2026, the faults we see most often on the bench, and whether it is worth keeping alive.

Specifications at a glance

Chip and hashboard architecture

The M21 belongs to MicroBT’s M2x generation, the 2019 family that also produced the M20, M20S and M21S. All of them run MicroBT’s own in-house SHA-256 silicon — what our firmware research catalogs as the K-series ASIC (internal chip family K-Gen-3, detected on stock firmware as chip ID 0x1920). Unlike Bitmain, which designs the BM13xx line and fabs at TSMC, MicroBT designs its hashing silicon and has it built at Samsung’s foundry. The M21 is a lower-output bin of that same generation: where the higher-tier M20S carries 105 chips per hashboard, the M21 trades chip count and clocks for a smaller 31 TH/s package and a gentler ~1,860 W draw.

The chips are wired in long series chains across each hashboard. An important accuracy note that trips up a lot of spec pages: voltage on these boards is regulated per power domain — a group of series-wired chips sharing one regulated rail — not per individual chip. There is no per-chip voltage knob on any production SHA-256 ASIC, MicroBT or Bitmain. Understanding that domain structure is the foundation of every board-level repair we do.

The control board: pure ARM, no FPGA

This is where MicroBT’s design philosophy diverges sharply from Bitmain’s. The M2x control board (the CB2 family) is built around an Allwinner H3 — a quad-core, 32-bit ARM Cortex-A7 SoC. There is no FPGA. The H3 talks to the hashboards directly over UART/SPI and runs MicroBT’s closed-source btminer daemon on a custom Linux build. Contrast that with an Antminer of the same era, where a Xilinx Zynq (an ARM core fused to an FPGA fabric clocked at 667 MHz) handles the midstate work. The practical upshot: MicroBT boards are simpler, with fewer exotic parts to fail, but their hashboards and control boards are not cross-compatible with Antminer hardware — different SoC, different ASIC protocol, different EEPROM format, different connectors.

Real-world power and efficiency

The 1,860 W on the label is the figure measured at the board, not the number on your power bill. Once you account for power-supply conversion losses (typically 5–7% on a unit of this vintage), expect a real wall draw closer to 1,950–2,000 W on a 200–240 V circuit. That works out to the quoted ~60 J/TH efficiency — and that single number is the whole story of the M21 in 2026.

Sixty joules per terahash put the M21 firmly in the legacy tier. For perspective, MicroBT’s current M70S does the same work at roughly 13.5 J/TH, so the M21 burns about four and a half times the energy per unit of hashrate. There is very little useful tuning headroom on this silicon: the K-series chips were not built for the deep autotuning that later 5nm parts enjoy, and any autotuner profile is calculated at runtime from live board behaviour, never a fixed preset. The most sensible “tuning” here is the other direction — a modest underclock to shave watts and lower fan noise when the unit is being run for heat rather than yield. If you want to understand the trade-offs across the whole fleet, our ASIC power profiles database lays out efficiency-versus-output curves by model.

The flip side of low efficiency is that the M21 is a very effective electric heater. Essentially all 1,900-odd watts become heat — about 6,346 BTU/h — and unlike a dumb resistive heater, that heat comes with a small Bitcoin reward attached. Ducted into a garage, workshop or grow space, the M21 turns money you would have spent on heating into hashrate.

Firmware compatibility

Stock firmware is MicroBT’s OpenWrt-based image running the proprietary btminer mining daemon and API. The Whatsminer API speaks a custom JSON-RPC over TCP port 4028 — not the CGMiner protocol Antminer and Avalon use, despite sharing the port number. The read API (summary, pools, devices) is enabled by default; write commands (pool changes, reboot, frequency targets) are disabled out of the box and must be unlocked with MicroBT’s WhatsMinerTool utility, which is also how firmware is normally flashed. SD-card recovery via PhoenixCard exists as a fallback when a unit will not take a network update.

Third-party firmware reality on the M2x generation is honest but thin. A small ecosystem of performance firmware grew up around the higher-volume S variants (the M20S and M21S), offering autotuning and modest overclocks; the base M21 sees far less coverage. Worth stating plainly so nobody wastes an afternoon: BraiinsOS is an Antminer-only project and was never ported to Whatsminer hardware, and the popular ePIC universal control-board swap targets the later M3x/M5x boards, not the 2019 CB2. For a broader primer on the trade-offs between stock, third-party and open firmware, see our guide to ASIC mining firmware types.

On the open-source side, D-Central’s own DCENT_OS effort is building toward multi-vendor Whatsminer support as part of a GPL-3.0 toolchain that is currently in closed beta (public beta is targeted for summer 2026). We are upfront that K-series support is early and we would rather under-promise here than oversell a work in progress — credit where it is due, MicroBT’s closed stack is mature and reliable, and replacing it is a long road.

Common faults and troubleshooting

After years of these on the bench, the M2x failure pattern is predictable:

• One hashboard dropping out — the unit reports a fraction of its rated hashrate because a board has fallen offline. Usually a dead chip or a broken domain in one of the series chains.
• Missing or “X” chips — the chain enumerates fewer ASICs than expected, pointing to a specific failed chip, a cracked solder joint, or a clock/reset issue partway down the chain.
• Fan and over-temperature errors — a stalled or out-of-spec fan trips thermal protection and shuts the unit down to protect the chips.
• PSU faults — the power supply on a six-year-old unit is a common wear-out item; symptoms range from no power-on to random restarts under load.
• Control-board corruption — a bad flash or aging eMMC on the Allwinner H3 board. MicroBT’s init routine self-checks its own binaries and will throw a checksum error code if the firmware image is damaged.

If your M21 is showing a specific symptom or code, start with our ASIC fault finder and error-code database, and cross-reference the front-panel status LED blink codes to narrow a no-hash unit down to power, control board, or hashboard before you open anything.

Repair and longevity

D-Central has run an in-house ASIC repair lab since 2016, and the M2x series is squarely within what we fix. Because the M21 is built from discrete, serviceable parts — chained chips on conventional hashboards, a removable PSU, and a standard Allwinner control board — it is a very repairable machine. We diagnose dead boards down to the failing chip or domain using a dedicated Whatsminer M20/M21 hashboard test fixture, then handle chip-level rework, domain repair, fan and PSU replacement, and control-board reflashing or swaps.

The economics matter on a unit this old: a repair only makes sense when the running cost still pencils out, so we will give you a straight answer about repair-versus-replace rather than fixing something that should be retired. Our general ASIC repair service covers the M21, and the board-level techniques carry over from our newer Whatsminer hashboard repair work. Kept clean, well-fed with airflow, and run within temperature, a healthy M21 will hash for years.

Who the M21 is for

Be clear-eyed about it. At ~60 J/TH, the M21 will not turn a profit against the network at typical North American power prices — the electricity to run it costs more than the Bitcoin it earns. It earns its keep in three situations:

• Heat-first mining — you need the BTU anyway, and recovering a few sats from heat you’d generate regardless is pure upside.
• Free or near-free power — flared gas, curtailed solar, hydro overflow, or any setting where the marginal cost of electricity is effectively zero.
• Learning and tinkering — a cheap, real ASIC to practice pool configuration, monitoring and repair on without risking a flagship.

If your goal is to learn rather than to heat a room, a far better starting point today is a single-chip open-source miner — see which Bitaxe to buy and our wider open-source miner directory; they sip power, run quietly on a desk, and teach the same fundamentals. If you are shopping for a machine that can actually mine at a profit, skip the legacy gear and compare current-generation hardware in our ASIC miner database or browse what we stock in Canada.

Generational context

The M21 marks the end of MicroBT’s 7nm chapter. The very next year the M30S series moved to Samsung 8nm and roughly halved the energy-per-terahash; the 2022 M50 line jumped to 5nm, and the M60 (2023) and M70 (2025) families pushed efficiency into the low-teens of J/TH. Seen against that arc, the M21 is a historical workhorse — reliable, repairable, and honest about what it is.

For live profitability math, full specifications and side-by-side comparisons against any other model, look up the MicroBT Whatsminer M21 in our ASIC miner database.